Wan 2.2 Notes - This model works very well with Wan 2.2. Output with plain wan 2.2, causvid v2, and lightx2v all seem to work well. Videos using Wan 2.2 are being posted and will be visible once Civitai's processor gets around to it. The LoRA settings you have used in the past should continue to work with Wan 2.2, just don't forget to apply this LoRA to both of the model -> ksampler pipelines to get the best results. After some experimentation, I found that you can mix CausVid and LightX2V in the two model pipelines. They are good at slightly different things. For my taste, I prefer the output I get with CausVid for the first high noise render, then LightX2V for the final low noise render. You can do your own experiments and decide what fits your tastes the best.

Edited - I revised my Wan 2.2 notes above because my original test videos had an error in the workflow which led me to some wrong assumptions. With a corrected workflow, the LoRA works extremely well at just about any resolution and configuration.

This clothing LoRA was inspired by the excellent Overhang - PDXL LoRA. It's an attempt to offer an alternative to the default "cropped" look offered by Wan which is very tight to the body. It's nearly impossible to prompt Wan to create a loose cropped shirt, so I decided to try and make a clothing model to help.

The keyword is "cr0pt0p", with zeroes in place of the letter O in case that's not clear in the Civitai font.

Because many of my training images were on the bustier side, use of this model will likely also result in a bigger bust for your video characters. The training data included t-shirts, hoodie sweatshirts, knitted sweaters, camisoles, and tank tops in different colors and materials so you should be able to prompt for a wide variety of clothing. The model will give you a nearly transparent top if you use the descriptor "sheer". For example, "She is wearing a pink sheer camisole."

This is my first clothing LoRA and I am gradually learning some lessons. I may later release another version of this model after tweaking my training captions. This model is a little too strong about changing the person, not just the shirt. If you add this model to another Wan character model, it will probably work the way you hope. At least, it works well with my BustySigrid model without changing her face, although it does boost her bust size some. As I said, there are some things I'd like to improve about this version. Still, I think it's good enough to release for people to play with.

As usual, please don't forget to post your Wan clips to this model page or tag this model as a resource when you upload your clip. That way I can see the scenes other people are making with the model. I hope you enjoy.